This is the first post on the Ligate Labs Mirror. A short letter on what we're working on, why now, and what to look for in the next twelve months.

A permissionless attestation chain. The pitch in one sentence: any human, app, or AI agent can publish proof of what they did and have it settled by a neutral third party, without trusting any single provider.

Schemas define the shape of the receipt. Attestor sets sign them under a threshold. The chain stores hashes and signatures, never the underlying content. That last bit matters more than it sounds, and I'll get to why.

Three apps ship on top, all on the same chain, all under one token:

• Themisra. Proof of Prompt. The first canonical schema. AI provenance for humans using AI tools.

• Iris. MCP server plus USD-billed relayer for autonomous AI agents. The pattern is "sponsored gas": agents call the API, we handle the chain side, customers pay in dollars. Open source MCP, SaaS margin.

• Kleidon. Four Web3 SaaS products on the same protocol. Passify, SkinsVault, TokenForge, MintMarket.

Anyone else can register their own schemas and attestor sets on the same protocol. We're not gatekeeping. The bet is that we win by being the default substrate, not by closing the gate behind us.

Sovereign SDK rollup on Celestia. Devnet target Q2 2026.

Three things are converging, and I think the timing is real, not a narrative I'm forcing.

AI generation hit regulatory scale. The EU AI Act, Article 50, demands that AI-generated content be cryptographically markable. Today's solutions are all loose. C2PA strips on re-upload. Statistical watermarks degrade through compression. Voluntary provider disclosures aren't auditable by anyone outside the provider. Regulators need a verifiable, privacy-preserving primitive that doesn't require trusting any single AI company. We wrote up exactly how this gets built in the latest essay.

Modular crypto matured. Sovereign SDK plus Celestia DA makes domain-specific rollups economically viable at AI inference volume. Our attestation fee is 0.001 LGT. A billion attestations a year becomes a million LGT in protocol fees. Compare that to Ethereum L1 gas, which would be structurally impossible at this volume. The infrastructure now exists. It didn't two years ago.

Agents need on-chain identity. Autonomous AI agents are signing transactions, calling APIs, moving capital. They need a settlement layer for "what did the agent actually do" that humans, regulators, and other agents can audit. Iris is the developer surface for that. Ten lines of integration. We're starting with this because it's the narrowest, most immediate use case, and it's where the first revenue will come from.

A few things that are non-negotiable for us, and a few we're still figuring out.

Permissionless before permissioned. Anyone can register a schema. Anyone can register an attestor set. We don't decide what gets attested. This is uncomfortable for a startup because it means we can't easily charge gatekeeping fees. We think it's worth it.

Privacy by default. The chain stores hashes and signatures, never plaintext. GDPR-compatible, regulator-compatible, citizen- compatible. This is the part I get most excited about. Attestation infrastructure that doesn't become surveillance infrastructure is not a contradiction in terms, it's a design choice.

Schema owner sets the trust model, not the protocol. A national authority, a standards body, an audit firm, a corporate consortium: each picks their own attestor set and threshold. We're the substrate. They're the policy.

Earn each version on the previous version's traction. v0 ships federated attestor sets. v1 adds slashable bonds and disputes. We're not pre-spending belief. If v0 doesn't get traction with real schemas and real attestors, v1 doesn't matter.

What we're still figuring out: how to bootstrap the first ten schemas without subsidizing them, what the right relationship is between protocol fees and schema-owner fees, how aggressively to push on the privacy-preserving prompt fingerprint design before we've stress-tested it adversarially. These are open questions. If you have strong opinions, send them.

• Devnet Q2 2026, late-2026 production-ready realistic.

• Iris paying customers as the early revenue signal.

• Design partners: 1 to 2 studios or SaaS teams piloting Themisra schemas.

• Closing pre-seed in parallel with infrastructure work.

If you're a regulator, an audit firm, an AI provider, a standards body, or a builder thinking about provenance at the protocol level, hello@ligate.io is the inbox. The whitepaper is the technical reference. Subscribe here for occasional updates as we ship.

— Stefan